On Thu, Feb 23, 2012 at 07:22:36PM +0000, Aidan Wong wrote: >That one node that recently had the running away memory leak was sending >253 metrics. I'm using unicast sending all metrics to a specific host >where I have configured the "udp_send_channel" with the "host" and "port" >attributes defined.
IIRC plugins are loaded once and then run within gmond's address space. so I guess plugins could be causing memory leaks. which plugins are you using? they alloc/free as they should? we haven't notived any leaks (certainly no serious leaks) across our ~1800 gmonds using 3.2.0, but we aren't sending that many metrics either - just using most of the standard stuff plus modified diskstat, cputemp python plugins, and with a bunch of other metrics spoof'd from chassis and switches (more cpu cycles for HPC job this way). we are using multicast. all except a few gmonds (not included below) are senders only. %CPU %MEM VSZ RSS COMMAND min 0.0 0.0 70972 1864 /usr/sbin/gmond median 0.0 0.0 70972 3392 /usr/sbin/gmond ave 0.0 0.0 70977 3240 /usr/sbin/gmond max 0.0 0.0 71104 4720 /usr/sbin/gmond those with the larger RSS have been rebooted recently and haven't yet had unused pages pushed out by vm pressure. cheers, robin -- Dr Robin Humble, HPC Systems Analyst, NCI National Facility ------------------------------------------------------------------------------ Virtualization & Cloud Management Using Capacity Planning Cloud computing makes use of virtualization - but cloud computing also focuses on allowing computing to be delivered as a service. http://www.accelacomm.com/jaw/sfnl/114/51521223/ _______________________________________________ Ganglia-general mailing list Ganglia-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/ganglia-general